The Net GHG Balance and Budget of the Permafrost Region (2000–2020) From Ecosystem Flux Upscaling

The northern permafrost region has been projected to shift from a net sink to a net source of carbon under global warming. However, estimates of the contemporary net greenhouse gas (GHG) balance and budgets of the permafrost region remain highly uncertain. Here, we construct the first comprehensive bottom‐up budgets of CO2, CH4, and N2O across the terrestrial permafrost region using databases of more than 1000 in situ flux measurements and a land cover‐based ecosystem flux upscaling approach for the period 2000–2020. Estimates indicate that the permafrost region emitted a mean annual flux of 12 (−606, 661) Tg CO2–C yr−1, 38 (22, 53) Tg CH4–C yr−1, and 0.67 (0.07, 1.3) Tg N2O–N yr−1 to the atmosphere throughout the period. Thus, the region was a net source of CH4 and N2O, while the CO2 balance was near neutral within its large uncertainties. Undisturbed terrestrial ecosystems had a CO2 sink of −340 (−836, 156) Tg CO2–C yr−1. Vertical emissions from fire disturbances and inland waters largely offset the sink in vegetated ecosystems. When including lateral fluxes for a complete GHG budget, the permafrost region was a net source of C and N, releasing 144 (−506, 826) Tg C yr−1 and 3 (2, 5) Tg N yr−1. Large uncertainty ranges in these estimates point to a need for further expansion of monitoring networks, continued data synthesis efforts, and better integration of field observations, remote sensing data, and ecosystem models to constrain the contemporary net GHG budgets of the permafrost region and track their future trajectory.


Introduction
The northern permafrost region covers up to 21 million km 2 of land in the Northern Hemisphere, of which ca. 70% (14 million km 2 ) is entirely underlain by permafrost (Obu et al., 2021)-ground that is at or below 0°C for at least two consecutive years.Unprecedented and amplified increases in air temperature in the Arctic (Rantanen et al., 2022) have strong impacts on the permafrost ground temperatures and extent (Biskaborn et al., 2019;Li et al., 2022), with future climate projections indicating a potential loss of permafrost extent of 4.0 ( 1.1 + 1.0, 1σ confidence interval) million km 2 for each °C of global temperature change (Chadburn et al., 2017).Consequences are already visible, as ground temperatures near the depth of zero annual amplitude in the continuous permafrost zone increased by 0.39 ± 0.15 °C between 2007 and 2016, reducing the permafrost extent by 7% between 1969 and 2018 (Biskaborn et al., 2019;Li et al., 2022).Changes in ground temperature expose substantial quantities of organic carbon (C), resulting in C degradation and atmospheric release of greenhouse gases (GHGs) such as carbon dioxide (CO 2 ), methane (CH 4 ), and nitrous oxide (N 2 O) from permafrost into the atmosphere (Chen et al., 2021;Natali et al., 2019;Schuur et al., 2009Schuur et al., , 2015;;Treat et al., 2018;Voigt et al., 2020a).This release of GHGs to the atmosphere could have a strong impact on the global carbon cycle as the upper three m of permafrost region soils are estimated to store 1,000 ± 200 Pg (1 Pg = 1,000 Tg) of soil organic carbon (Hugelius et al., 2014;Mishra et al., 2021) and 55 Pg of soil nitrogen (N) (Palmtag et al., 2022).Deeper deposits store an additional 400-1,000 Pg C, making the permafrost region the largest terrestrial carbon and nitrogen pool on Earth (Schuur et al., 2022).These soil C and N have been accumulating over thousands of years due to limited microbial decomposition at low temperatures and water-logged conditions, leading to long-term accumulation of organic matter and incorporation into permafrost (Tarnocai et al., 2009).Upon thaw-that can occur gradually or abruptly-permafrost landscapes are changing, impacting their hydrology and biogeochemical cycling, creating a potentially significant feedback to the global climate (Schuur et al., 2008(Schuur et al., , 2015(Schuur et al., , 2022)).The release of GHGs from permafrost has the potential to accelerate global climate warming, known as the "permafrost carbon feedback" (Burke et al., 2017(Burke et al., , 2022;;Schuur et al., 2015).While longer growing seasons, increased CO 2 concentrations, and additional nutrient release from thawing permafrost may lead to increased vegetation productivity and partly offset the release of permafrost GHGs (Koven et al., 2015;Liu, Kuhn, et al., 2022;López-Blanco et al., 2022;McGuire et al., 2018;Schuur & Mack, 2018), other processes such as disturbances cause rapid shifts to landscape structure (Schuur et al., 2008(Schuur et al., , 2011) ) and might accelerate the release of GHGs into the atmosphere.
Although presumably crucial for the global carbon cycle, the role of the northern permafrost region in the global carbon budget is unknown.Existing estimates of terrestrial GHG exchange from land cover-based or machine learning-based ecosystem vertical flux upscaling identify the northern permafrost terrestrial ecosystem as a net sink of CO 2 ( 181 Tg CO 2 -C yr 1 , Virkkala et al., 2021) and a net source of N 2 O (0.14-1.03Tg N 2 O-N yr 1 , Voigt et al., 2020a), although large uncertainties remain.The northern permafrost region GHG budgets remain poorly constrained as our understanding of the GHG balance of this region has been hampered by low data availability (both temporal and spatial) and a heterogeneous landscape that is complex to map accurately.Watts et al. (2023) show that in northern high latitude, the Net Ecosystem Carbon Budget is reduced by ca.7% when inland waters (e.g., lakes, ponds, streams, and rivers)-known to be significant emitters of CO 2 and CH 4 (Cole et al., 2007;Serikova et al., 2018;Stanley et al., 2016;Thornton et al., 2016;Wik et al., 2016)-are included, and by ca.30% when emissions from inland waters and fires are considered.However, no study has yet included inland waters and disturbances to constrain the GHG budget of the permafrost region and provide an overall net GHG balance.
Here we fill this gap and present comprehensive budgets of GHGs (CO 2 , CH 4 , and N 2 O) by key permafrost land cover types over the period 2000-2020 across the northern permafrost region using a single flux upscaling approach for all three GHGs.We include most relevant ecosystem components, that is, terrestrial ecosystems, inland waters, geological fluxes, lateral fluxes, and fire fluxes.However, we exclude direct anthropogenic emissions from the combustion of fossil fuels and land use change.The total carbon and nitrogen budgets that we provide for the region represent storage change budgets.This permafrost regional budget is part of the REgional Carbon Cycle Assessment and Processes-2 (RECCAP2) project of the Global Carbon Project that aims to collect and integrate regional GHGs budgets for 12 land regions and 5 ocean basins covering all global lands and oceans (Ciais et al., 2022; https://www.globalcarbonproject.org/ reccap/).Comparisons of GHG budgets using this upscaling flux approach and budgets based on atmospheric inversion models and terrestrial process-based models are discussed in Hugelius et al. (2023).

Study Area
The spatial extent of permafrost defined in this study includes areas within the northern permafrost region as defined in Obu et al. (2021) and restricted to the Boreal Arctic Wetlands and Lakes Data set area that had the key land cover classes for our upscaling (Boreal-Arctic Wetland and Lake Dataset (BAWLD), Olefeldt et al., 2021) (Figure 1).As a consequence, the BAWLD-RECCAP2 permafrost region does not include large areas underlain by permafrost in Central Asia and the Tibetan plateau (shown in Figure S1 in Supporting Information S1).The BAWLD-RECCAP2 permafrost region considered in this study is 18.5 million km 2 .All flux estimates were scaled to the BAWLD-RECCAP2 permafrost region (hereafter permafrost region).The study area overlaps several other RECCAP2 regions (Ciais et al., 2022), and no specific effort to harmonize the budgets presented here with the RECCAP2 budgets of those regions is made in this paper.

GHG Budgets From Ecosystem Flux Upscaling
Data-driven ecosystem flux upscaling of GHG budgets for a reference time period of 2000-2020 was calculated by summing up flux budgets from terrestrial ecosystems, inland waters, lateral fluxes, fire emissions, and geological fluxes.To calculate the total net regional GHG flux (F x ), we used the following equation: where F x is the annual permafrost region gas flux for the GHG species of interest x, A j is the area of each land cover class j (Figure 2, Table S1), and F jx is the land cover average GHG flux density for species x (Table S1).
We used existing synthesis databases and upscaled gridded data products published in the past 5 years to estimate annual and growing season mean fluxes per land cover type.All budget numbers are presented as the weight of C and N (i.e., CO 2 -C, CH 4 -C and N 2 O-N yr 1 ) and not as the weight of GHG molecules.Budgets are reported as mean fluxes with uncertainties providing the 95% confidence intervals (CI) in Tg C or N.

GHG Fluxes From Terrestrial Land Cover Types
The land cover classification used for the analysis was adapted from the BAWLD set land cover (Olefeldt et al., 2021).The BAWLD land cover classes are distinguished based on moisture regime, nutrient/pH regime, organic-soil depth, hydrodynamics, and the presence or absence of permafrost (Olefeldt et al., 2021).To match the observational GHG flux data sets, we simplified the nine terrestrial land cover classes in BAWLD into five: Boreal Forests, Non-permafrost Wetlands, Dry Tundra, Tundra Wetlands, and Permafrost Bogs (Figure 2).Classes were defined as: • Non-permafrost Wetlands include permafrost free bogs, fens, and marshes with no near-surface permafrost (see Canadian Wetland Classification system).• Boreal Forests are forested ecosystems with non-wetland soils.Coniferous trees are dominant, but the class also includes deciduous trees in warmer climates and/or certain landscape positions.Boreal forest ecosystems may have permafrost or be permafrost free.• Permafrost Bogs are ecosystems with near surface permafrost and thick surface peat layers (>40 cm).This includes palsas, peat plateaus, and the elevated portions of high-and low-center polygonal permafrost bogs.They typically have ombrotrophic conditions that cause nutrient-poor conditions.The vegetation is dominated by lichens, Sphagnum mosses, woody shrubs, and sometimes sparse coniferous forest.• Dry Tundra include treeless ecosystems (both lowland arctic and alpine tundra) dominated by graminoid or shrub vegetation.Dry Tundra ecosystems generally have near-surface permafrost.Dry Tundra is differentiated from Permafrost Bogs by their thinner organic soil (<40 cm), and from Tundra Wetlands by their drained soils (average water table position >5 cm below soil surface).• Tundra Wetlands are treeless ecosystems with near surface permafrost and saturated to inundated conditions for large parts of the year.Tundra Wetlands can both be mineral (<40 cm peat) or have peat (>40 cm peat).They are distinguished from Dry Tundra and Permafrost Bogs by being wetter and having more dynamic hydrology.Tundra Wetlands include areas that can be classified as tundra fen wetlands in the Canadian Wetland Classification System.
BAWLD was developed specifically to fit the needs of CH 4 upscaling and was dependent on available CH 4 flux data for different land cover classes.This choice of land cover classes was made after assessing the type of sites in three flux databases of CO 2 , CH 4 , and N 2 O used for the upscaling (see description below), ensuring that there was sufficient data for each class and that the merging was the most parsimonious grouping that allowed us to estimate each GHG balance for each class.To match the observational GHGs data sets (see section below), we simplified wetland landcover classes for N 2 O and CO 2 fluxes based on available flux data, related information on landcover classes, and differences in mechanistic drivers.For example, for N 2 O fluxes, wetland classes were originally more coarsely separated into "peatlands" (>30 cm of organic layer) and "wetlands" (peatlands other than peatlands).Thus, N 2 O Wetland fluxes were applied to the Tundra wetland and Marsh total areas, while N 2 O Peatland fluxes were applied to the Bog, Fen, and Permafrost Bog total areas.For CH 4 upscaling, we kept the five wetland classes included in the work by Kuhn et al. (2021).Due to a lack of flux data, rocklands and glaciers were not included in the classification.The area of each land cover class (A j ) in km 2 across the permafrost region is shown in Figure 2 and detailed in Table S1.
The land cover mean GHG flux (F jx ) was obtained for each of the five terrestrial land cover classes after homogenizing and analyzing three comprehensive GHG flux data sets:  (Kim & Tanaka, 2003;Köster, Köster, Berninger, Heinonsalo, & Pumpanen, 2018;Matson et al., 2009;Morishita et al., 2007;Schiller & Hastie, 1996;Simpson et al., 1997;Ullah et al., 2009) (Treat et al., 2018), except for boreal forests were we assumed growing season emissions accounted for 100% of annual emissions as the sites averaged net CH 4 growing season uptake and available data for winter season fractions only covers CH 4 -emitting ecosystems (Treat et al., 2018).Therefore, our Boreal Forest annual estimate should be considered conservative.Annual N 2 O fluxes were estimated assuming that growing season emissions accounted for 50% of annual emissions as reported in Voigt et al. (2020a).For all three GHGs, only sites with no record of large-scale upland hillslope abrupt thaw disturbance in the metadata were included in the flux estimates to avoid double-counting emissions from upland hillslope abrupt thaw (see methodology for disturbances).However, although scarce, we included other disturbed sites in our CO 2 estimates to account for ecosystem CO 2 losses following disturbances and their different successional stages (e.g., four sites reporting thermokarst; Virkkala et al., 2022).Sites from the above-mentioned GHG flux data sets were classified into one of the five terrestrial land cover classes using the metadata provided in each of the data sets.
While the focus of this study is the period 2000-2020, we include all in situ measurements obtained between 1991 and 2020 in order to overcome the limited amount of flux measurements in some of the ecosystems and therefore ensure adequate spatial representation of ecosystem fluxes.A separate analysis of decadal CO 2 fluxes from 1991 to 2020 revealed no differences, suggesting that the extension of the time series to 1991 does not impact our findings (Table S3).
More details on how ecosystem flux upscaling was performed (Text S1 in Supporting Information S1) as well as the temporal coverage of these data sets can be found in supplementary material (Figure S2 in Supporting Information S1, Table S2).

Vertical GHG Fluxes From Inland Waters
Similar to the method used to calculate GHG emissions from terrestrial land cover types, GHG fluxes from inland waters were calculated by upscaling mean GHG fluxes from lakes and rivers (see below) using the estimated surface area of these aquatic classes from the BAWLD classification (Olefeldt et al., 2021) adjusted to the study region (see supplementary Table S1 for estimated aerial extent of inland waters).

Vertical GHG Fluxes From Rivers
Atmospheric riverine GHG fluxes were calculated in different ways for each GHG, depending on available source data, and when possible scaled across the region using riverine area from the permafrost region (0.12 × 10 6 km 2 ), as reported in BAWLD.
Estimates of river and stream CO 2 flux were calculated from gridded monthly flux data estimated by Liu (2021; https://doi.org/10.5061/dryad.d7wm37pz9;Dryad) from river water dissolved CO 2 pressure and gas transfer velocity.This data is delivered as unprojected global grids with a 0.0083°resolution (which is ca 1 × 0.2 km pixels in the high Arctic).The global grids were clipped to the extent of BAWLD and then reprojected to an equal area grid at 100 × 100 m resolution.Vertical riverine fluxes were scaled across the riverine area in the domain BAWLD (0.12 10 6 km 2 ).Ice-free season river CO 2 emissions were calculated based on the mean-annual flux reported for the Arctic region by Liu, Kimball, et al. (2022).The river CO 2 flux reported by Liu, Kimball, et al. (2022) of 1,750 g C m 2 yr 1 incorporates the number of ice-free days; therefore, we scaled across the domain by multiplying the flux by the riverine area.To estimate total annual river CO 2 emissions-including emissions released after spring ice-out-we assumed that ice-out emissions accounted for 17% of annual emissions (Denfeld et al., 2018; consistent with the approach by Liu, Kimball, et al., 2022).Riverine CH 4 emissions were determined using the mean CH 4 diffusive flux reported in the MethDB (Stanley et al., 2016).Stanley et al. (2016) found that diffusive CH 4 emissions did not statistically differ across latitudes and scaled global river CH 4 emissions using one mean value.Given the limited number of reported CH 4 fluxes for rivers in the Arctic, we used the same approach as Stanley et al. (2016) and applied a global mean diffusive flux of 135 mg CH 4 m 2 d 1 to the river area.Because there are few studies that measure CH 4 emissions upon ice-out, we applied for CH 4 the same estimate as for CO 2 , where 17% of annual fluxes occur during the ice-free period (Denfeld et al., 2018; consistent with the approach by Liu, Kuhn, et al., 2022).Ebullition was not included for river CH 4 emission estimates due to few available measurements in the literature for this region (Stanley et al., 2016).
Estimates of river N 2 O flux were derived from gridded annual N 2 O flux estimated by a mechanistic mass balance model developed globally for inland waters by Maavara et al. (2019).These data were reprojected from an original 0.5°unprojected grid to an equal area grid at 1 km resolution and clipped to the BAWLD extent.As the original lake and river surface area was not known, no correction of the inland water surface area was made.
Uncertainties for river GHG balance were determined using the standard error and coefficient of variance reported by Liu, Kuhn, et al. (2022), Stanley et al. (2016) and Maavara et al. (2019), respectively, for CO 2 , CH 4 , and N 2 O.

Vertical GHG Fluxes From Lakes
CH 4 fluxes (diffusion and ebullition) were extracted from the BAWLD-CH4 aquatic ecosystem data set and classified based on classes (yedoma lakes, peatland ponds, and glacial/post-glacial organic poor lakes and ponds) and sizes, from large (>10 km 2 ) to midsize (0.1-10 km 2 ) to small lakes (<0.1 km 2 ) (Kuhn et al., 2021a; total area = 1.255 10 6 km 2 ; Table S4).As in Olefeldt et al. (2021) any area -regardless of its size -which is likely to be inundated >50% of the growing season period (long term average) is considered part of the lake land cover classes.Ice-free days were determined based on averages of reported ice-free days for each lake type and this information was used to determine ice-free season fluxes (supplementary Table S1).In addition to ice-free emissions, spring ice-out emissions (i.e., winter contribution) were considered to be 23% of the annual total (Wik et al., 2016).
Estimated lake CO 2 fluxes were compiled from multiple available sources based on a literature search made in May 2022 (Humborg et al., 2010;Karlsson et al., 2013;Kortelainen et al., 2006;Pelletier et al., 2014;Rasilo et al., 2015;Rocher-Ros et al., 2017;Sepulveda-Jauregui et al., 2015) and are summarized in Table S5).The studies report lake CO 2 fluxes as mean flux values for various binned lake surface areas.We took these averages and grouped them by the lake size classes included in BAWLD (<0.1, 0.1-10, >10 km 2 ).We found no statistical differences in fluxes between the size groups and thus used one mean lake CO 2 flux to scale across the year and the region (315 ± 196 mg C m 2 d 1 ).We applied the same number of ice-free days used to scale lake CH 4 emissions (ice-free days reported in the literature for each lake class).
To estimate lake fluxes of N 2 O, gridded global data of annual flux from Lauerwald et al. (2019) were used.This estimate is based on the nitrous oxide (N 2 O) emission model developed by Maavara et al. (2019) and the HydroLAKES database and was reprojected from an original 0.5°unprojected grid to an equal area grid at 1 km resolution and clipped to the BAWLD extent.As the original lake and river surface area was not known, no correction of the inland water surface area was made.Uncertainties for lake N 2 O were determined using the coefficient of variance reported for regions north of 50°latitude in Lauerwald et al. (2019).

Disturbances-GHG Fluxes From Fires and Abrupt Thaw
Monthly GHG fire emissions were extracted for the study region from the Global Fire Emission Database version 4s for the period 1997-2016 and its Beta version for the years 2017-2020 (GFED; van der Werf et al., 2017).The GFED4s estimates of burned areas are based on remote sensing data at a spatial resolution of 0.25°(van der Werf et al., 2017).GHG emissions in the GFED4s are derived from the multiplication of burned area and fuel consumption per unit burned area, the latter being the product of modeled fuel loads per unit area and combustion completeness.For our purpose, we extracted mean annual GHG emissions from burned areas using the GFED4s for the period 2000-2016 and the GFED4s Beta for the period 2017-2020.To validate the use of the GFED4s Beta for the later period we compared the emissions obtained with the ones by van Wees et al. ( 2022) (Figure S3 in Supporting Information S1).
Localized, but widespread, disturbances associated with abrupt thaw are thought to contribute significantly to GHG emissions from permafrost (Abbott & Jones, 2015;Holloway et al., 2020;Marushchak et al., 2021;Runge et al., 2022;Turetsky et al., 2020;Walker et al., 2019;Yang et al., 2018).Abrupt thaw includes thawing processes that affect permafrost soils in periods of days to several years (Grosse et al., 2011), and is typically associated with thermokarst and thermoerosion processes that lead to the formation of hillslope erosional features (thaw slumps, thermo-erosion gullies and active layer detachments), thermokarst lakes, and thermokarst wetlands (i.e., collapse scar bogs and fens).We report abrupt thaw areas and derived annual CO 2 and CH 4 emissions using the inventorybased abrupt thaw model by Turetsky et al. (2020), in which atmospheric emissions are estimated for three generalized types of abrupt thaw terrains: mineral-rich lowlands, upland hillslopes, and organic-rich wetlands.In the abrupt thaw model, abrupt thaw areas are based on synthesized field observations and remote sensing measurements.GHG emissions from abrupt thaw were synthesized for each ecosystem state within each abrupt thaw type from the literature (ca.20 published papers).The abrupt thaw model was initialized for a historical assessment period  to provide the model with a spin up and prevent the regional carbon fluxes starting at zero at the beginning of the dynamic measurement period.Thaw rates were generally in equilibrium with succession and recovery of surface permafrost during this initialization period.Changes in the area of each successional state were tracked over time by multiplying initial starting areas by transition rates.Estimates of abrupt thaw GHG emissions following the historical assessment period were performed by increasing rates of abrupt thaw through time.This increase in thaw rate was prescribed to follow the average output of "permafrostenabled" land surface models, all of which were forced by atmospheric climate anomalies from the Community Climate System Model 4 Earth system model under an RCP8.5 projection.For our purpose, we ran the abrupt thaw model for the period 2000-2020, extracted cumulative CO 2 and CH 4 emissions from active and stabilized abrupt thaw features, and derived annual fluxes for each abrupt thaw terrain for the time period 2000-2020.We used the reported uncertainty ranges of ±40% on the upland hillslope areas, ±30% on the mineral-rich lowland areas, and ±35% on the organic-rich wetland areas, as in Turetsky et al. (2020).Additional details on the inventory model can be found in Turetsky et al. (2020).Since GHG data sets that we used for ecosystem upscaling partly account for abrupt thaw and to prevent double counting of GHG fluxes, CO 2 and CH 4 fluxes from abrupt thaw were added as a sub-flux (not added to the total) of terrestrial and inland water land cover fluxes and their Global Biogeochemical Cycles 10.1029/2023GB007953 contribution to the total GHG budget is discussed.Due to the lack of in situ observations of abrupt thaw impacts on N 2 O fluxes in the used data sets, no N 2 O balance is presented for abrupt thaw.

Lateral Fluxes and Geological Emissions
Lateral C and N fluxes from riverine transport and coastal erosion (i.e., dissolved organic carbon (DOC) and dissolved organic nitrogen (DON) losses from the permafrost region to the ocean) are taken from Terhaar et al. (2021), representative for all land north of 60°N.They estimated riverine lateral fluxes for the six largest Arctic rivers (Mackenzie, Yukon, Kolyma, Lena, Ob, Yenisei) from the Arctic Great River Observatory data set and extrapolated them to the entire Arctic catchment.Emissions from coastal erosion were calculated by multiplying spatially resolved estimates of coastal erosion rates by estimates of C content in coastal soils provided by Lantuit et al. (2012).
Estimates of geological emissions of CH 4 (from subsurface fossil hydrocarbon reservoirs) are taken from an upscaled circumpolar permafrost region estimate for gas seeps along permafrost boundaries and lake beds made by Walter Anthony et al. (2012).We note that there is some risk of double counting such fluxes, especially in sites where eddy covariance flux towers may have unknowingly been placed close to seeps of geological CH 4 emissions.No separate estimates of geological emission for CO 2 or N 2 O are available for the permafrost region.

Net GHG Balance From Terrestrial Land Cover Types
Terrestrial ecosystems represent a decadal-scale sink for CO 2 and source for CH 4 and N 2 O (Table 1, Figure 3).The mean annual CO 2 flux was a net sink, but could not be distinguished from CO 2 neutral when the 95% confidence interval was considered ( 339.6 ( 835.5, 156.3)Tg CO 2 -C yr 1 ).The broad uncertainty interval can be attributed both to the large natural variability in CO 2 fluxes across sites and to the heterogeneity of ecosystem types included in each of the land cover classes defined in the BAWLD classification.Boreal Forests and Nonpermafrost Wetlands were CO 2 sinks ( 270.3 and 69.4 Tg CO 2 -C yr 1 , respectively), while Tundra Wetlands and Permafrost Bogs were close to neutral ( 2.7 and 0.05 Tg CO 2 -C yr 1 , respectively).Dry Tundra was the only ecosystem type classified as an annual ecosystem CO 2 source (2.9 Tg CO 2 -C yr 1 ), but the very broad uncertainty range ( 147.6, 153.5 Tg CO 2 -C yr 1 ) indicates low confidence in the sign of this flux.Terrestrial ecosystems were overall a net sink of CO 2 during the growing season ( 1,611 ( 2148, 1,074) Tg CO 2 -C gs 1 ), with the strongest sink in the boreal forest ( 1,034 ( 1,305, 763) Tg CO 2 -C gs 1 ) (Table 2).
Annual vertical terrestrial CO 2 balance has been reported for high-latitudes in recent papers using different upscaling approaches.While closely related due to overlap in flux data, a higher NEE uptake is reported by both Virkkala et al. (2021) and Watts et al. (2023) ( 419 (95% CI of 559 to 189) Tg CO 2 -C yr 1 and 601 (standard error of ±1138) Tg CO 2 -C yr 1 , respectively).However, the estimated NEE uptakes for the permafrost region solely are weaker, with an uptake of 181 ( 305, 32) Tg CO 2 -C yr 1 and 230 (±22) Tg CO 2 -C yr 1 , respectively).The difference between the later NEE uptakes and our results relates to the subset of data included in the analyses (exclusively eddy covariance tower fluxes in Watts et al., 2023), the different years covered in the analyses (Virkkala et al., 2021: 1990-2015, Watts et al., 2023: 2003-2015), the different spatial extents, and the upscaling approach applied (Arctic Terrestrial Carbon Flux Model (TCFM-Arctic) in Watts et al., 2023, andstatistical upscaling in Virkkala et al., 2021).Both of these studies as well as the previous RECCAP synthesis (1990( -2006( , McGuire et al., 2012) ) report the tundra as a weak CO 2 sink ( 13 ( 81, 62); 16 (±84-270); and 16 ( 42, 10) Tg CO 2 -C yr 1 , respectively) although they also show that annual tundra CO 2 balance cannot be distinguished from CO 2 neutral when taking into account the uncertainty range.Dry Tundra CO 2 balance was also identified as a source of 10 ( 27, 47) Tg CO 2 -C yr 1 in McGuire et al. (2012).
Our estimated annual net CH 4 source of 25.6 (14.7, 36.4)Tg CH 4 -C yr 1 from terrestrial ecosystems (Table 1) was largely driven by emissions from Non-permafrost Wetlands (20.6 (14.3,26.9)Tg CH 4 -C yr 1 ).As in Treat et al. (2018), Non-permafrost Wetlands emitted more than Tundra Wetlands.Annual CH 4 flux estimates for Tundra Wetlands (3.3 (2.7, 3.9) Tg CH 4 yr 1 ) and Dry Tundra (2.1 ( 0.4, 4.5 Tg CH 4 -C yr 1 ) were in the lower range from the previous estimates provided by McGuire et al. (2012), in which the tundra was estimated to release 11 (0, 22) Tg CH 4 -C yr 1 (between 1990 and 2006).Our growing season CH 4 budget was a source of 16 (8.6,23.3) Tg CH 4 -C gs 1 (Table 2), with non-permafrost Wetlands contributing 83%.All terrestrial ecosystems except Boreal Forests were net CH 4 emitters.Boreal Forests were a net sink of CH 4 ( 1.1 ( 2.3, 0.0) Tg CH 4 -C gs 1 ).This net CH 4 uptake from Boreal Forests was shown in a synthesis of site-level and plot-level CH 4 on the order of 1.1 mg CH 4 m 2 d 1 (Kuhn et al., 2021).This is largely driven by net CH 4 oxidation in the aerobic soils, many of which are covered by forest (non-Sphagnum) mosses.Moreover, recent investigations of the regulatory mechanisms of CH 4 oxidation in Boreal Forests show evidence that SOC constitutes an important variable that governs the forest CH 4 sink by providing alternative carbon compounds for methanotrophs and enhancing their activity by stimulating growth (Lee et al., 2023).The net sink of atmospheric CH 4 in well-drained soils is widely recognized in global budgets (e.g., Saunois et al., 2019) as well as in earlier process-based studies (Whalen et al., 1992).Our CH 4 annual balance was lower than those estimated for the northern high latitude wetlands (>45°N) at 31, 32, and 35 Tg CH 4 C yr 1 (depending on wetland distribution maps) by Peltola et al. (2019) and 38 Tg CH 4 C yr 1 by Watts et al. (2023).However, our CH 4 growing season balance estimate was higher than the balance based on 93 observations presented in Treat et al. (2018) except for the Tundra Wetlands where they remain within the same range.Despite their large spatial coverage, the Dry Tundra was a small of CH 4 during the growing season (1.4 ( 0.3, 2.9) Tg CH 4 -C gs 1 ), although the low end of the CI suggests that it could remain a sink.More measurements from these drier ecosystems are needed, as also recent studies indicate that tundra soils, in particular well-drained uplands, could be important CH 4 sinks (D'Imperio et al., 2023;Oh et al., 2020;Voigt et al., 2023).Note.Negative GHG emissions represent an uptake while positive emissions represent a release.GHG emissions from terrestrial ecosystems are reported as mean fluxes with 2.5 and 97.5% confidence intervals (CI).GHG emissions from inland waters and fires are reported with 5 and 95% CI.GHG emissions from abrupt thaw are reported with ±40% uncertainty range.Bold values were to emphasize the columns with the mean values.a These fluxes are estimated using the abrupt thaw model from Turetsky et al. (2020) and are considered as additive to the total for these categories (to avoid double counting of fluxes).b Includes CO 2 , CH 4 , and lateral fluxes.Forests were the second largest N 2 O source (0.14 ( 0.01, 0.30) Tg N 2 O-N yr 1 ) due to their large area, although their fluxes per unit area were small (Table S6, 52.43 μg N 2 O m 2 d 1 ).Although they occupy a small portion of the landscape (5%), Permafrost Bogs were the largest N 2 O emitters per unit area (Table S6, 645.14 μg N 2 O m 2 d 1 ) and their contribution to the regional N 2 O balance was 18%.The estimate for Permafrost Bogs includes emissions from barren peat surfaces, where vascular plants are absent -surfaces previously identified as N 2 O hot spots in the Arctic due to ideal conditions for N 2 O production (Gil et al., 2022;Marushchak et al., 2011;Repo et al., 2009).A challenge remains regarding the mapping of Permafrost Bogs and barren ground and integration within land cover classifications.Therefore, we did not differentiate between vegetated and non-vegetated Permafrost Bog areas when upscaling.N 2 O emissions from Tundra Wetlands were negligible (0.01 (0.00, 0.02) Tg N 2 O-N yr 1 ), which can be explained by the lack of nitrate supply as an N 2 O precursor under reduced conditions and reduction of N 2 O to N 2 during denitrification when the water table is high (Butterbach-Bahl & Dannenmann, 2011;Voigt et al., 2017).Recent observations not included in the N 2 O review data set (Voigt et al., 2020a) show that wetlands may also function as net N 2 O sinks in the Arctic (Schulze et al., 2023).

Net GHG Balance From Inland Waters
Inland aquatic ecosystems were a net source of CO 2 (230.6 (132.4, 359.8)Tg CO 2 -C yr 1 ), CH 4 (9.4 (4.5, 13.1) Tg CH 4 -C yr 1 ), and N 2 O (0.0019 (0.0008, 0.0029) Tg N 2 O-N yr 1 ).Rivers annually emit 164.4 (107.3,222.5)Tg CO 2 -C yr 1 , 2.3 (1.6, 2.9) Tg CH 4 -C yr 1 and 0.0006 (0.0004, 0.0008) Tg N 2 O-N yr 1 to the atmosphere.These high riverine fluxes are due to their supersaturation in CO 2 as they receive and degass CO 2 derived from adjacent soils.To our knowledge, there are no specific annual estimates of riverine GHGs for the permafrost region to compare our estimates; however, when compared to emissions from high latitudes, our CH 4 emissions for rivers are within the lower range of published estimates (0.3-7.5 Tg CH 4 -C yr 1 ) (Thornton et al., 2016).
Our annual lake CH 4 emission estimate is lower than previous estimates reported by Wik et al. (2016) (12.4 (7.3, 25.7) Tg CH 4 -C yr 1 ) and Matthews et al. (2020) (13.8-17.7 Tg CH 4 -C yr 1 ).This is partly related to the difference in lake classifications where in this study lakes were separated by both types and size categories, whereas these previous estimates separated the lakes by type alone-although domain sizes differ slightly.The largest source of lake CH 4 emissions was from small peatland lakes (∼30% of lakes emissions, Table S4), which are dominant in the peat-rich regions of the Hudson Bay Lowlands in Canada and the West Siberian Lowlands in western Russia (Olefeldt et al., 2021).However, the areas of small lakes estimated by BAWLD are among the most uncertain of the land cover classes (Olefeldt et al., 2021) due to limited spatial data used for lakes and great flux variability among small lakes across the domain (Muster et al., 2019).Our mean lake and river CO 2 emission estimates for the permafrost region constitute ∼12% of reported global annual CO 2 emissions for lakes (Holgerson & Raymond, 2016) and rivers (Liu, Kimball, et al., 2022).We note that there is a substantial lack of CH 4 flux data for Boreal-Arctic lakes (Stanley et al., 2016), making our estimates highly uncertain.While there is no estimate of N 2 O emissions from arctic lakes, Kortelainen et al. (2020) estimated boreal lakes N 2 O emissions at 0.029 (0.026, 0.032) Tg N 2 O-N yr 1 .

Net GHG Balance From Disturbances
Fires within the study region affected 1.1 × 10 6 km 2 during the period 2000-2020.On average, fires impacted 0.06 million km 2 annually, emitting 121 (96.7, 145.3)Tg CO 2 -C yr 1 , 1.8 (1.4,2.1) Tg CH 4 -C yr 1 , and 0.12 (0.10, 0.15) Tg N 2 O-N yr 1 .Ninety percent of the annually burned area was in the boreal biome, contributing to more than 92% of the permafrost region fire GHG emissions (Table 1).Fire CO 2 emissions offset a third of the CO 2 uptake from terrestrial ecosystems, while CH 4 and N 2 O emissions from fires represent 7% and 21% of the CH 4 and N 2 O emitted by terrestrial ecosystems, respectively.Our fire emission estimates mainly reflect direct emissions from combustion.There is also a component of increased growth during post-fire recovery, which we do not explicitly account for.However, it is indirectly accounted for as many of the in situ flux data were collected from previously burned ecosystems (which drives up the mean land cover flux).Our fire carbon emission estimate for boreal ecosystems (CO 2 and CH 4 , 123 TgC yr 1 ) is slightly lower than that of 142 Tg CO 2 -C yr 1 previously reported by Veraverbeke et al. (2021).Using GFED4s data, our estimate might underestimate fire CO 2 emissions, as shown in Potter et al. (2022), where GFED4s emissions were 36% lower than those obtained using the ABoVE-FED data-driven product.

Global Biogeochemical Cycles
10.1029/2023GB007953 The total area affected by active and stabilized abrupt thaw between 2000 and 2020 was estimated to be 1.2 × 10 6 km 2 (0.43 × 10 6 in lowlands, 0.01 × 10 6 in uplands, and 0.72 × 10 6 in wetlands), accounting for ca.7% of the permafrost region (Table 1).Altogether, areas affected by abrupt thaw were net emitters of 31 (21, 42) Tg CO 2 -C yr 1 and 31 (20, 42) Tg CH 4 -C yr 1 (Table 1, details in Table S7).CO 2 and CH 4 emissions from wetland abrupt thaw were the largest (Table 1).GHG estimates from abrupt thaw were not directly included in the permafrost GHG balance as it was not possible to know how much were already accounted for in the GHG balance from terrestrial upscaling.However, the impact of abrupt thaw processes on C cycling in the permafrost region is large, and it is projected that it will contribute nearly as much as gradual thaw to future radiative forcing from permafrost thaw (Turetsky et al., 2020).
Taking into account all the above mentioned budget components, the total C (including atmospheric CO 2 , CH 4 , and lateral fluxes) budget for the permafrost region between 2000 and 2020 was estimated to 144.3 ( 505.6, 825.8)Tg C yr 1 .Atmospheric CO 2 contributed ca.8% to the total C released from the region, while atmospheric CH 4 contributed 26.5%.The total N budget for the permafrost region was 3.3 (2.0, 4.8) Tg N yr 1 .Most of the (76%) the N released was through lateral fluxes, with coastal erosion releasing 30% of the total N from the region.Atmospheric N 2 O from inland waters was negligible, whereas atmospheric N 2 O from terrestrial ecosystems represented 17% of the total N released in the permafrost region.Atmospheric N 2 O losses due to fires represented 4% of the N in the permafrost region.

Limitations in the Number of Observations
A major challenge in the representation of GHG exchange in high-latitude and remote environments relates to limitations in spatial representation, length and quality of observational time series (Pallandt et al., 2022;Virkkala et al., 2018).The synthesis data sets used here to estimate GHG fluxes are the most comprehensive ones currently available and have been significantly growing during the past decade.However, more observations covering the full annual cycles are still needed to improve the representativeness of heterogeneous and underrepresented landscapes and climatic conditions.Specifically, more observations from the Dry Tundra land cover class are needed to verify its GHG sink-source status and from ecosystems experiencing disturbances such as abrupt thaw.CH 4 flux measurements are limited in Boreal Forests, and N 2 O flux measurements are scarce for all terrestrial and aquatic ecosystems.Across all the GHG fluxes, measurements in environments with low fluxes are also important to avoid biasing our understanding to hotspot regions.Limitations related to the number of flux measurements could be overcome by increasing in situ and laboratory experiments.This would improve the process-based understanding of fluxes and their response to changes in temperature, moisture, permafrost thaw and other disturbances.Improvements in the reporting of measurements and metadata should be prioritized for a better integration of available data, especially to address reporting of net-zero or negative fluxes.Difficulties in measuring small exchange rates can be overcome by using new technologies based on portable, high-precision laser instruments (e.g., D'Imperio et al., 2017;Juncher Jørgensen et al., 2015).Very recently, such portable high-precision instruments are also becoming available for N 2 O, opening possibilities for more numerous and accurate N 2 O flux estimates, including the capture of N 2 O uptake.
N 2 O flux measurements from inland waters are still scarce and ice-out estimates are often missing for CH 4 fluxes.Moreover, seasonally inundated water bodies are not well represented although they might contribute substantially to the release of GHGs in short periods of time.
Estimates of high latitude lateral fluxes of C and N are fairly well constrained in comparison to land-atmosphere GHG fluxes.However, available estimates are provided for the major six largest arctic rivers that represent 50% Global Biogeochemical Cycles 10.1029/2023GB007953 of the total area covered by rivers (Speetjens et al., 2023).Although smaller catchments are highly abundant, estimates of GHG fluxes are not well constrained for the permafrost region.Improving this understanding will allow lateral flux integration of these smaller catchments in the main estimates of lateral fluxes from inland waters.

Limitations Related to the Land Cover Classification
Differences in GHG fluxes among land cover classes are large.Therefore, it is crucial to obtain their representation correctly to improve land cover-based GHG flux upscaling.To date, there is no accurate land cover classification of permafrost landscapes (both dry and wet) at a circumpolar scale.We used the BAWLD land cover classification (Olefeldt et al., 2021) S6), making the interpretation of the flux estimates difficult.
Emissions from small water bodies (<0.1 km 2 ) globally represent important inland water CO 2 and CH 4 fluxes (Holgerson & Raymond, 2016) and even more at high latitudes.Although accounted for in this study, emissions from small water bodies are quite uncertain as they are difficult to map at a large scale due to their high temporal and spatial variability.Small ponds and lakes can be temporary and their size can vary depending on the amount of precipitation after snowmelt; they expand much in wet years and after snowmelt and can often disappear in dry years or late in summer.Improving the spatial and temporal resolution of the products used to map inland waters would benefit the representation of small water bodies, which would resolve a critical source of uncertainty in calculating GHG exchange.

Limited Understanding of the Impact of Disturbances on the GHG Budget
As ecosystems go through disturbance cycles, there are both losses and gains of C and N to ecosystems.It is unclear how well post-disturbance dynamics, for example, post-fire regrowth, is captured in our ecosystem flux upscaling.Our emissions from fires consider direct GHG emissions but not the indirect and longer-term soil emissions resulting from fire-induced ground thaw.Although carbon losses might be offset by shifts in species composition (Mack et al., 2021;Randerson et al., 2006;Ueyama et al., 2019), fires can also initiate further permafrost thaw and degradation (Genet et al., 2013;Gibson et al., 2018;Jafarov et al., 2013).As such, fires can trigger shifts in the landscapes, impacting biogeochemical cycling (Abbott & Jones, 2015;Bouskill et al., 2022;Hermesdorf et al., 2022;Köster, Köster, Berninger, Prokushkin, et al., 2018;Marushchak et al., 2021;Randerson et al., 2006;Ullah et al., 2009;Voigt et al., 2017;Wilkerson et al., 2019).Improving our understanding of landscape transitions due to fire will help constrain the contribution of disturbances to the GHG balance in the region.Next GHG budgets for the permafrost region will need to call for new data sets of fire emissions that account for post-fire recovery processes.
The spatial extent and GHG emissions from abrupt thaw disturbances remain poorly constrained due to a lack of available data (Turetsky et al., 2020).Flux measurements from abrupt thaw are still scarce and thus their reported flux estimate should be interpreted carefully.Improving the number of in situ measurements from abrupt thaw disturbances and consistent reporting should be a key to understanding the impact of abrupt thaw on permafrost GHG budgets.Transition rates (from active to stabilized abrupt thaw feature) need to be further understood and systematic mapping of abrupt thaw areas remain to be improved to better constrain emissions from abrupt thaw.N 2 O emissions from abrupt thaw were not included in this study due to the small number of observations reported in the literature and little understanding on the impact of abrupt thaw on emissions N 2 O.However, it has been shown that such disturbances frequently cause N 2 O emission hotspots (Voigt et al., 2020a).Two recent studies using a terrestrial ecosystem model simulate enhanced gaseous N losses from thawing permafrost (Lacroix et al., 2022;Yuan et al., 2023).Another study showed that uptake of atmospheric N 2 O in peat plateaus and thermokarst bogs increased with soil temperature and soil moisture following disturbances (Schulze et al., 2023).Local hydrology will determine whether the site will turn into a source of N 2 O after thaw as high emissions can occur at intermediate moisture conditions in N rich soils (Marushchak et al., 2021), but a transition to wetland would promote denitrification with N 2 as the final product and prevent Global Biogeochemical Cycles 10.1029/2023GB007953 N 2 O release (Butterbach-Bahl & Dannenmann, 2011;Voigt et al., 2017) or even cause or enhance net N 2 O uptake (Schulze et al., 2023).
As our understanding of processes leading to GHG release through abrupt thaw is constantly improving, future permafrost GHG budgets will be able to better integrate both atmospheric and lateral fluxes from abrupt thaw.Thus far, the abrupt thaw model (Turetsky et al., 2020) does not consider lateral fluxes from abrupt thaw.While we might capture these losses through our lateral fluxes, future budgets should allow measuring the fraction of what is lost due to abrupt thaw.Other disturbances including anthropogenic disturbances (e.g., clear cutting and logging) have not been estimated in this study.Future budgets could aim at constraining the impact of these disturbances on the permafrost GHG budget.

Conclusions
Using a land cover-based ecosystem flux upscaling approach (including fluxes from terrestrial ecosystems, inland water, disturbances and geological fluxes), the permafrost region was identified as an annual source of GHGs between 2000 and 2020.The region emitted 12 ( 606, 661) Tg CO 2 -C yr 1 (mean and 95% confidence interval range used hereafter), 38 (22, 53) Tg CH 4 -C yr 1 , and 0.67 (0.07, 1.3) Tg N 2 O-N yr 1 to the atmosphere.The region was thus a net source of CH 4 and N 2 O.For CO 2 , although the 20-year mean is a net source, the uncertainty range remains large, extending from a large sink to an even larger source of CO 2 and therefore challenging the calculation of the net flux sign.We suggest that terrestrial ecosystems were likely an ecosystem CO 2 sink, but emissions from disturbances and inland waters offset this flux, making the full CO 2 budget largely indistinguishable from zero (neutral).The total C (including atmospheric CO 2 , CH 4 , and lateral fluxes) and N budgets for the permafrost region were estimated to 144 ( 506, 826) Tg C yr 1 and 3 (2, 5) Tg N yr 1 .
Estimates of river and stream CO 2 flux were calculated from gridded monthly flux data estimated by Liu (2021; https://doi.org/10.5061/dryad.d7wm37pz9;Dryad).Estimates of river N 2 O flux were derived from gridded annual N 2 O flux estimated by a mechanistic mass balance model developed globally for inland waters by Maavara et al. (2019).CH4 fluxes (diffusion and ebullition) were extracted from the BAWLD-CH4 aquatic ecosystem data set (Kuhn et al., 2021a).Estimated lake CO 2 fluxes were compiled from multiple available sources based on a literature search made in May 2022 (Humborg et al., 2010;Karlsson et al., 2013;Kortelaine et al., 2006;Pelletier et al., 2014;Rasilo et al., 2015;Rocher-Ros et al., 2017;Sepulveda-Jauregui et al., 2015) and are summarized in Table S5).To estimate lake fluxes of N 2 O, gridded global data of annual flux from Lauerwald et al. (2019) were used.Monthly GHG fire emissions were extracted for the study region from the Global Fire Emission Database version 4s for the period 1997-2016 and its Beta version for the years 2017-2020 (GFED; van der Werf et al., 2017).We report abrupt thaw areas and derived annual CO 2 and CH 4 emissions using the inventory-based abrupt thaw model by Turetsky et al. (2020).Lateral C and N fluxes from riverine transport and coastal erosion (i.e., dissolved organic carbon and dissolved organic nitrogen losses from the permafrost region to the ocean) are taken from Terhaar et al. (2021).Estimates of geological emissions of CH 4 (from subsurface fossil hydrocarbon reservoirs) are taken from an upscaled circumpolar permafrost region estimate for gas seeps along permafrost boundaries and lake beds made by Walter Anthony et al. (2012).

Figure 1 .
Figure 1.The BAWLD-RECCAP2 region defined as the northern permafrost extent (data from Obu et al., 2021) restricted to the Boreal Arctic Wetlands and Lakes Data set area (BAWLD, Olefeldt et al., 2021).The Figure overlays the extents of the Boreal Forests and the Tundra on the BAWLD-RECCAP2 region as well as the observation sites used for upscaling GHGs.

Figure 2 .
Figure 2. Circumpolar percentage coverage of the five adapted Boreal-Arctic Wetland and Lake Dataset (BAWLD) terrestrial land cover types (Boreal Forests, Non-permafrost Wetlands, Permafrost Bogs, Dry Tundra, and Tundra Wetlands) used for ecosystem-based upscaling of greenhouse gas flux budgets in this study.Note that these maps show the distributions across the full BAWLD domain as presented by Olefeldt et al. (2021), not the more limited extent of the RECCAP2 permafrost BAWLD domain used in this study.
annual balance estimate of 0.55 ( 0.03, 1.1) Tg N 2 O-N yr 1 (Table1) suggests that terrestrial ecosystems were a N 2 O source, although the uncertainty range around N 2 O fluxes extends from a small sink to a larger source.These high uncertainties partly relate to the limited number of observations of N 2 O fluxes (47 sites and 91 observations), which only include growing-season observations.Our estimated annual N 2 O balance is within the range of the one previously reported byVoigt et al. (2020a) (0.14-1.03Tg N 2 O-N yr 1 median-meanbased estimate).In our study, Dry Tundra was the largest N 2 O source (0.23 (0.04, 0.42) Tg N 2 O-N yr 1 ).Boreal
mogenizing and analyzing three comprehensive GHG flux data sets: Virkkala et al. (2022) for CO 2 fluxes; Kuhn et al. (2021a) for CH 4 fluxes; and Voigt et al. (2020a, 2020b) for N 2 O fluxes.Additional data were extracted from the literature for Boreal Forest N 2 O fluxes

Table 1
Greenhouse gas (GHGs-CO 2 , CH 4 , and N 2 O) Balance and Total C and N Budgets for the Permafrost Region Based on Ecosystem Flux Upscaling

Table 2
Growing Season (gs) Emissions of Greenhouse Gas (GHGs-CO 2 , CH 4 , and N 2 O) From Terrestrial Ecosystems in the Permafrost Region Note.GHG emissions are reported as mean fluxes with 2.5 and 97.5% confidence intervals (CI).a Non-permafrost Wetlands include fens, bogs, and marshes.Due to lack of data, N 2 O fluxes for non-permafrost Wetlands, fluxes are assumed to be equal to those of Tundra Wetlands.
in which land cover classes were defined to enable upscaling of CH 4 fluxes at large spatial scales.While very relevant to facilitate large-scale mapping of CH 4 fluxes, it lacks sufficient classes to allow separation among groups of dryer ecosystems that might have large variability in CO 2 or N 2 O fluxes.This is the case for the Dry Tundra and Boreal Forest classes that comprise a mosaic of ecosystems with different vegetation types.This results in a large uncertainty range in the class flux estimate of the Dry Tundra (see Table1, Table